powerpc/cpuidle: Set CPUIDLE_FLAG_POLLING for snooze state

During the comparative study of cpuidle governors, it is noticed that the
menu governor does not select CEDE state in some scenarios even though when
the sleep duration of the CPU exceeds the target residency of the CEDE idle
state this is because the CPU exits the snooze "polling" state when snooze
time limit is reached in the snooze_loop(), which is not a real wake up
and it just means that the polling state selection was not adequate.

cpuidle governors rely on CPUIDLE_FLAG_POLLING flag to be set for the
polling states to handle the condition mentioned above.

Hence, set the CPUIDLE_FLAG_POLLING flag for snooze state (polling state)
in powerpc arch to make the cpuidle governor work as expected.

Reference Commits:

- Timeout enabled for snooze state:
  commit 78eaa10f02
  ("cpuidle: powernv/pseries: Auto-promotion of snooze to deeper idle state")

- commit dc2251bf98
  ("cpuidle: Eliminate the CPUIDLE_DRIVER_STATE_START symbol")

- Fix wakeup stats in governor for polling states
  commit 5f26bdceb9
  ("cpuidle: menu: Fix wakeup statistics updates for polling state")

Signed-off-by: Aboorva Devarajan <aboorvad@linux.vnet.ibm.com>
Tested-by: Vishal Chourasia <vishalc@linux.vnet.ibm.com>
Reviewed-by: Vaidyanathan Srinivasan <svaidy@linux.ibm.com>
Reviewed-by: Vishal Chourasia <vishalc@linux.vnet.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Link: https://lore.kernel.org/r/20221114145611.37669-1-aboorvad@linux.vnet.ibm.com
This commit is contained in:
Aboorva Devarajan 2022-11-14 20:26:11 +05:30 committed by Michael Ellerman
parent 3ae7c96dd5
commit 5ddcc03a07
2 changed files with 10 additions and 3 deletions

View File

@ -76,6 +76,7 @@ static int snooze_loop(struct cpuidle_device *dev,
local_irq_enable();
snooze_exit_time = get_tb() + get_snooze_timeout(dev, drv, index);
dev->poll_time_limit = false;
ppc64_runlatch_off();
HMT_very_low();
while (!need_resched()) {
@ -86,6 +87,7 @@ static int snooze_loop(struct cpuidle_device *dev,
* cleared to order subsequent test of need_resched().
*/
clear_thread_flag(TIF_POLLING_NRFLAG);
dev->poll_time_limit = true;
smp_mb();
break;
}
@ -155,7 +157,8 @@ static struct cpuidle_state powernv_states[CPUIDLE_STATE_MAX] = {
.desc = "snooze",
.exit_latency = 0,
.target_residency = 0,
.enter = snooze_loop },
.enter = snooze_loop,
.flags = CPUIDLE_FLAG_POLLING },
};
static int powernv_cpuidle_cpu_online(unsigned int cpu)

View File

@ -44,6 +44,7 @@ static int snooze_loop(struct cpuidle_device *dev,
pseries_idle_prolog();
local_irq_enable();
snooze_exit_time = get_tb() + snooze_timeout;
dev->poll_time_limit = false;
while (!need_resched()) {
HMT_low();
@ -54,6 +55,7 @@ static int snooze_loop(struct cpuidle_device *dev,
* loop anyway. Require a barrier after polling is
* cleared to order subsequent test of need_resched().
*/
dev->poll_time_limit = true;
clear_thread_flag(TIF_POLLING_NRFLAG);
smp_mb();
break;
@ -268,7 +270,8 @@ static struct cpuidle_state dedicated_states[NR_DEDICATED_STATES] = {
.desc = "snooze",
.exit_latency = 0,
.target_residency = 0,
.enter = &snooze_loop },
.enter = &snooze_loop,
.flags = CPUIDLE_FLAG_POLLING },
{ /* CEDE */
.name = "CEDE",
.desc = "CEDE",
@ -286,7 +289,8 @@ static struct cpuidle_state shared_states[] = {
.desc = "snooze",
.exit_latency = 0,
.target_residency = 0,
.enter = &snooze_loop },
.enter = &snooze_loop,
.flags = CPUIDLE_FLAG_POLLING },
{ /* Shared Cede */
.name = "Shared Cede",
.desc = "Shared Cede",